Use your own AI (MCP)

Already live in Claude, Cursor or another AI tool? On Scale and Enterprise plans, connect it directly to your workspace over the Model Context Protocol — bring your own AI, keep the governance.

What it is

MCP (the Model Context Protocol) is the open standard AI clients use to call external tools. Your workspace exposes a curated tool surface — reading and writing content, reading code, checking deploys, image utilities — at https://mcp.agenticwebteam.com/mcp. Any MCP-capable client can use it: Claude Code, Claude Desktop, Cursor, VS Code and more. Your conversation stays in your client, with your AI provider; only the tool calls reach us.

The point is what doesn't change: every call runs under the same role-based permissions, draft/publish rules, approval requirements, audit log and metering as the portal assistant. A writer's key can't publish, whichever model is driving.

Connecting a client

Chat apps (claude.ai, Claude Desktop, Microsoft Copilot): no keys involved. Add a custom connector with the URL https://mcp.agenticwebteam.com/mcp and the app walks you through OAuth — you sign in to your workspace (Microsoft sign-on works here too), see exactly what you're granting and to which app, and approve. Disconnect any time from the portal's Integrations page.

Developer tools (Claude Code, Cursor, VS Code): on the portal's Integrations page, create an MCP access key — it acts with your permissions, is shown once, and can be revoked at any time (each teammate creates their own). Then add the server with the key as a bearer token. Claude Code, for example:

Then just ask your AI — “list our blog posts”, “draft a news post about X as a draft on staging” — and it works your site through the governed tools.

Governance & visibility

• Multi-site, one connection: a connection covers every workspace your email holds a seat in. You pick a default when connecting; “update the hero on the Acme site” just works — the AI names the workspace per call (list_workspaces shows them), and each workspace applies its own plan, permissions and audit trail.
• Permissions: connections are per user and carry that user's role in each workspace. Permission denials are final — the tools never work around them.
• Audit: every write lands in the workspace audit log, marked as coming via MCP.
• Activity: admins see every external-AI session on the Integrations page — who connected, which client (Claude, Cursor…), and the full list of calls it made, with outcomes. Activity expires with your data-retention window.
• Revocation: revoking a key cuts access within a minute; leaving a plan that includes MCP disables all keys automatically.

How it's billed

Your own AI does the thinking on your existing AI subscription — we don't charge model tokens for MCP work. Each tool call draws a small number of credits from your plan's pool instead (reads are cheapest; content writes and image operations a little more), and appears in Usage & Billing like any other work.